Simple implementation of caching at depth that should be useful for most use cases

73dabd8905c9eecf60c1aeedabd9bcfe745a2975

Honey Tree <kalakanlogs@proton.me>

default/config.yaml+7 -0 Viewed

   # (e.g {{random}} macro or lorebooks not as in-chat injections).
   # Otherwise, you'll just waste money on cache misses.
   enableSystemPromptCache: false
 +  # Enables caching of the message history at depth (if supported).
 +  # https://docs.anthropic.com/en/docs/build-with-claude/prompt-caching
 +  # -- IMPORTANT! --
 +  # Use with caution. Behavior may be unpredictable and no guarantees can or will be made.
 +  # Set to an integer to specify the desired depth. 0 (which does NOT include the prefill)
 +  # should be ideal for most use cases.
 +  cachingAtDepth: false
 # -- SERVER PLUGIN CONFIGURATION --
 enableServerPlugins: false

src/endpoints/backends/chat-completions.js+18 -1 Viewed

     const apiKey = request.body.reverse_proxy ? request.body.proxy_password : readSecret(request.user.directories, SECRET_KEYS.CLAUDE);
     const divider = '-'.repeat(process.stdout.columns);
     const enableSystemPromptCache = getConfigValue('claude.enableSystemPromptCache', false) && request.body.model.startsWith('claude-3');
 +    let cachingAtDepth = getConfigValue('claude.cachingAtDepth', false) && request.body.model.startsWith('claude-3');
     if (!apiKey) {
         console.log(color.red(`Claude API key is missing.\n${divider}`));

                 requestBody.tools[requestBody.tools.length - 1]['cache_control'] = { type: 'ephemeral' };
+            }
+        }
 -        if (enableSystemPromptCache) {
++
 +        if (cachingAtDepth !== false) {
 +            // There are extremely few scenarios in which caching the prefill is a good idea, it mostly just breaks everything
 +            const messageCount = convertedPrompt.messages.length;
 +            cachingAtDepth += convertedPrompt.messages[messageCount - 1].role === 'assistant' ? 1 : 0;
++
 +            if (messageCount - 1 - cachingAtDepth >= 0) {
 +                convertedPrompt.messages[messageCount - 1 - cachingAtDepth]['cache_control'] = { type: 'ephemeral' };
 +            }
++
 +            if (messageCount - 1 - cachingAtDepth - 2 >= 0) {
 +                convertedPrompt.messages[messageCount - 1 - cachingAtDepth - 2]['cache_control'] = { type: 'ephemeral' };
 +            }
 +        }
++
 +        if (enableSystemPromptCache || cachingAtDepth !== false) {
             additionalHeaders['anthropic-beta'] = 'prompt-caching-2024-07-31';
+        }
++
         console.log('Claude request:', requestBody);
         const generateResponse = await fetch(apiUrl + '/messages', {